DeepSeek-R1: an open reasoning model
Open-sourced under MIT, with reasoning on par with o1 and its full thinking visible.
MITlicense, free for commercial use and distillation
Why it matters
Its R1-Zero variant showed that reasoning can emerge from pure reinforcement learning. It topped the US App Store and shook the industry.
Costs and controversy
On January 27, Nvidia fell about 17% in a single day, wiping out nearly $600B of market value.
Built on
Led to
Qwen3Kimi K2DeepSeek-V4R1 in Nature
Source: arxiv.org ↗